Choose a model

The model picker sits above the composer. It lists every model the workspace knows about, grouped by the provider that offers it.

What is in the list

Two sources, combined:

• Tincograph — the built-in provider, always present.
• Anything you have configured in Settings, under the display name you gave it. See Bring your own model key.

Internally each entry is providerId/modelId, which is why two providers can both offer a model called llama3.1 without colliding.

The default is a placeholder

The default is Tincograph Pro, and Tincograph Pro is a mock. It runs in your browser, sends nothing anywhere, and answers by repeating what you typed back at you. Tincograph Fast is the same mock with a different label.

They exist so the workspace is usable before you have configured anything — every screen, the streaming, the stop button and the artifact extraction all work against it. They are not models and they will not answer a question.

If the replies look like an echo, that is what you are talking to. Add a provider and pick one of its models.

The picker sets the model for the next chat

A chat is stamped with a model when it is created and keeps it for the whole conversation. The picker sets what the next new chat will use.

So changing the picker while a chat is open does not move that conversation to another model, even though the button at the top now reads differently. To use a different model on the same question, pick it and start a new chat.

If a chat refers to a provider you have since removed from Settings, the workspace falls back to the built-in provider rather than failing to send. An unexpected echo after editing your provider list is that fallback.

No model is locked to a plan

Every plan has the AI feature enabled, including Free. There is no “this model needs an upgrade” state, because the limit is not the model — it is the shared credit allowance every model draws from. A short reply through our servers costs one credit whichever model produced it, and a long reasoning answer costs three.

That is a deliberate choice rather than an oversight: pricing a chat by which model answered it means the price changes under you when a default changes.

What a model failure looks like

A provider error is appended to the reply in the conversation, with the status code and the first part of the provider’s own message. The commonest causes are a wrong baseUrl, a key the provider has revoked, and a model id that provider does not offer. All three are in your Settings rather than on our side.